Low-Latency Gesture Recognition From Spatial Filtering of Single-Element Ultrasound Signals

Abstract

A-mode ultrasound (US) has been applied to detect morphological changes of skeletal muscles for gesture recognition in the human-machine interface (HMI). Reduced sensor number and latency are highly desirable in many scenarios of HMI. With a single US sensor, this paper investigated employing spatial filters on one frame of US echoes to enhance the performance of low latency gesture recognition. Common spatial pattern on depth (CSPD), adapted from common spatial pattern, was proposed to improve the separation properties of a single US frame for different hand gestures. The performance of CSPD was compared with two commonly used methods, and the results showed a significant improvement in recognition accuracy of a thirteen-class classification task. The filter coefficients were visualized to provide insights into the filtering properties of CSPD. The outcome of this work demonstrated the feasibility of employing only one frame of a single US sensor to recognize gestures with ultra-low latency, facilitating its usability in HMI applications.